Video platforms once ranked material largely on how many people clicked, and now rank primarily on how long they stayed. The change followed directly from where the revenue comes from.
Advertising inventory is measured in time
A platform sells opportunities to show advertising, and those opportunities are generated while a viewer is watching rather than at the moment they click.
A video producing a click and thirty seconds of viewing creates almost no inventory. One producing fifteen minutes creates several placements.
Optimising for clicks therefore optimises the wrong quantity, and the correction was to rank on the measure that actually produces revenue.
Click optimisation produced predictable failures
When ranking rewarded clicks, the efficient strategy was a title and image promising more than the video delivered, since the payoff arrived before the disappointment.
That degraded the experience while the numbers improved, because the metric registered the arrival and not the immediate departure.
Counting time watched closes the gap, since a viewer who leaves quickly now reduces the video's standing rather than leaving it unchanged.
Session behaviour matters beyond the single video
Platforms also weigh what happens after a video ends, because a viewer who continues watching produces more inventory than one who leaves.
A video that satisfies and hands the viewer onward is therefore more valuable than one that satisfies and ends the session, even at identical length.
This is why endings, recommendations and series structure receive attention out of proportion to the time they occupy in a video.
Length became a lever, then stopped being one
The immediate creator response was to make videos longer, since total watch time rises with duration even when engagement falls.
Platforms adjusted by weighing the proportion watched alongside the absolute total, which penalises padding while still rewarding genuinely long viewing.
The result is that length helps only where it is sustained, and an unnecessarily long video now performs worse than the shorter version of itself.
The distortions that remain
Subjects that are quick to answer are structurally disadvantaged, because the honest video is short and short videos accumulate less time.
Creators respond by bundling several small answers into one piece, which serves the metric and serves a searcher wanting one of them poorly.
Short-form feeds were partly a response to this, creating a separate surface where brevity is not penalised, measured on completion and repeat viewing instead.